Complex models may have model components distributed over a network and generally require significant execution times. The field of parallel and distributed simulation has grown over the past fifteen years to accommodate the need of simulating the complex models using a distributed versus sequential method. In particular, asynchronous parallel discrete event simulation (PDES) has been widely studied, and yet we envision greater acceptance of this methodology as more readers are exposed to PDES introductions that carefully integrate real-world applications. With this in mind, we present two key methodologies (con- servative and optimistic) which have been adopted as solutions to PDES systems. We discuss PDES terminology and methodology under the umbrella of the personal communications services application.

NOTE: This is a preliminary release of an article accepted by the ACM Transactions on Modeling and Computer Simulation. The definitive version is currently in production at ACM and, when released, will supersede this version. c○1998 by the Association for Computing Machinery, Inc. Permission to make digital/hard copy of all or part of this material without fee for personal or classroom use provided that the copies are not made or distributed for profit or commercial advantage, the ACM copyright/server notice, the title of the publication, and its date appear, and notice is given that copying is by permission of the ACM, Inc. To copy otherwise, to republish, to post on servers, or to redistribute to lists requires prior specific permission and/or a fee. Permissions may be requested from

Simulation optimization can be defined as the process of finding the best input variable values from among all possibilities without explicitly evaluating each possibility. The objective of simulation optimization is to minimize the resources spent while maximizing the information obtained in a simulation experiment. The purpose of this paper is to review the area of simulation optimization. A critical review of the methods employed and applications developed in this relatively new area are presented and notable successes are highlighted. Simulation optimization software tools are discussed. The intended audience is simulation practitioners and theoreticians as well as beginners in the field of simulation.

Exciting new opportunities for efficient simulation of complex stochastic systems are emerging with the development of parallel computers with many processors. In this paper we describe an application of a new distributed-event approach for speeding up a single long simulation run to study the transient behavior of a large non-Markovian network of queues. In particular, we implemented the parallel-prefix-based algorithm of Greenberg, Lubachevsky and Mitrani (1991) on the 8192-processor CM-2 Connection machine to simulate the departure times D(k, n) of the k th customer from the n th queue in a long series of single-server queues. Each queue has unlimited waiting space and the first-in first-out discipline; the service times of all the customers at all the queues are i.i.d. with a general distribution; the system starts out with k customers in the first queue and all other queues empty. Glynn and Whitt (1991) established limit theorems for this model, but unfortunately very little c...

We discuss the simulation of M replications of a uniformizable Markov chain simultaneously and in parallel (the so-called parallel replicated approach). Distributed implementation on a number of processors and parallel SIMD implementation on massively parallel computers are described. We investigate various ways of inducing correlation across replications in order to reduce the variance of estimators obtained from the M replications. In particular, we consider the adaptation of Stratified Sampling, Latin Hypercube Sampling, and Rotation Sampling to our setting. These algorithms can be used in conjunction with the Standard Clock simulation of uniformized chains at distinct parameter values and can potentially sharpen multiple comparisons of systems in that setting. Our investigation is primarily motivated by this consideration. 1 Introduction Parallel/distributed approaches for discrete event simulation may be categorized as singlerun and multiple-run methods. The majority of studies ...

The simulation of high--speed telecommunication systems such as ATM (Asynchronous Transfer Mode) networks has generally required excessively long run times. This paper reviews alternative approaches using parallelism to speed up simulations of discrete event systems, and telecommunication networks in particular. Subsequently, a new simulation method is introduced for the fast parallel simulation of a common network element, namely, a work--conserving finite capacity statistical multiplexer of bursty ON/OFF sources arriving on input links of equal peak rate. The primary performance measure of interest is the cell loss ratio, due to buffer overflows. The proposed method is based on two principal techniques: (1) the derivation of low--level (cell level) statistics from a higher level (burst level) simulation and (2) parallel execution of the burst level simulation program. For the latter, a time--division parallel simulation method is used where simulations operating at different interva...

Parallelizing a problem by statically assigning a fixed number of subproblems to each processor is very popular due to its simplicity and low communication overhead. In many cases it can be proved to be sufficient to randomly assign O(log N) subproblems to each of N processors in order to smooth out load imbalance due to varying subproblem sizes. However, this is not true for the tree structured computations considered in this paper because the subproblem sizes get less uniform when the number of generated subproblems is increased. Even under moderate assumptions, a polynomial number of subproblems needs to be assigned to each processor. Still, for machines with slow communication and applications with good splitting functions, a carefully designed randomized static load balancer can be a competitive alternative to dynamic load balancing schemes. The results also help to explain the impact of static load balancing as an initialization method for a subsequent dynamic load balancing phas...

Concurrent computing on networked collections of computer systems is rapidly evolving into a viable technology that is attractive from the economic, performance, and availability perspectives. Several software infrastructures that support such heterogeneous network-based concurrent computing have evolved, and are in use for productionquality high-performance computing. In this paper, we describe such a system, and present our experiences with its use for massively concurrent computing in the application domain of polymer physics. The application involves stochastic simulation of polymer chains for measuring scale-invariant phenomena at critical disorder. The parallelization is achieved through the EcliPSe toolkit, and conducted on a flexible, treestructured virtual machine made up of arbitrary and heterogeneous computing nodes dispersed across the country. These nodes cooperate to perform the simulation and pool results together in real time at a central node which initiates the parall...

Discrete-event stochastic simulation is one of the most commonly used tools for performance modeling and evaluation. Parallel/distributed simulation enables a simulation program to execute on a computing system containing multiple processors and aims in reducing the model's execution time. Three basic types of execution mechanisms have appeared. The first two (the conservative and the optimistic approach) aim in partitioning the simulation model into a number of sub-models, also called logical processes (LPs). Their emphasis, lies on the specification of the appropriate synchronization, deadlock handling and/or memory management algorithms. The third approach (known as the time parallel approach or simply as Multiple Replications in Parallel Time Streams), aims in overcoming the need for sufficiently long runs in steady-state stochastic simulations, by executing multiple replications of the entire model in a parallel fashion. This work, presents a fast parallel OpenMP based implementation, for multivariate queuing network simulations. The simulation results are statistically processed, by applying the classical regenerative method under the Lavenberg & Sauer sequential analysis procedure. The first experimental results indicate significant speedups accompanied by acceptable confidence interval coverage.

In recent years, encouraged by today's fast workstations and by software systems designed to transform workstation clusters into parallel programming environments, network of workstations have been increasingly used as computational engines. Networked workstations, however, are not ideal replacements for supercomputers, because of the low interconnection capacity provided by current local area networks. In this paper, we present an application using the EcliPSe toolkit, a system for replication-based parallel processing in heterogeneous environments. Although primarily designed for replicative applications (that generally do not require large amounts of communication), EcliPSe can be used for more general forms of message-passing parallel processing. We describe the use of the toolkit in the parallelization of a cluster labeling algorithm. The algorithm is designed so that it uses some of EcliPSe's features to reduce communication overhead, making the algorithm suitable for execution o...